Utilization of apricot seed in (co-)combustion of lignite coal blends: Numeric optimization, empirical modeling and uncertainty estimation

• Published in: Fuel (Guildford) 2018-03, Vol.216, p.190-198
• Main Authors: Buyukada, Musa, Aydogmus, Ercan
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Ash; Bayesian analysis; Bayesian approach; Co-combustion; Coal; Combustion; Empirical analysis; Environmental aspects; Gases; Lignite; Mathematical models; Multiple nonlinear regression; Neural networks; Particle swarm optimization; Regression analysis; Response surface methodology; Seeds; Statistical analysis; Statistical models; Thermogravimetric analysis; Uncertainty
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2017.12.028

[Display omitted] •Thermochemical conversion of AS was investigated by LC-based (co-)combustion.•Lower ash amount of AS demonstrated that harmful gas release could be decreased.•ANN (99.9%) predicted MLP (%) better than MNLR (95.9%).•PSO was found more successful than RSM in optimization of operating conditions.•Bayesian approach was found quite effective for uncertainty estimation (±7.52%). Utilization of apricot seed (AS) in lignite coal (LC)-based (co-)combustion process was aimed in the present study considering the apricot production capacity of Turkey. By this way, an alternative and also ecofriendly way was suggested for coal-based energy production plants located in Turkey. This purpose was tested by thermogravimetric analyses to demonstrate the advantageous sides of AS in reduction of ash amount and also environmental aspects based on harmful gases. The other important contributors of present study was the comparison of both statistical modeling and numeric optimization techniques for maximization of mass loss percentage (MLP, %) in response to (co-)combustion process. For this purpose, multiple non-linear regression (MNLR) and artificial neural network (ANN) models as data-driven modeling techniques, and response surface methodology (RSM) and particle swarm optimization (PSO) as numeric optimization approaches were utilized. Results demonstrated the accuracy of ANN and PSO in prediction of MLP (%) and optimization of operating conditions of (co-)combustion of AS and LC, respectively. Finally, Bayesian approach was applied to the best-fit MNLR model to identify the uncertainties in predictors of proposed model. Bayesian was found quite effective in identification of uncertainties that were not possible to be captured through deterministic ways.